YOLOv12 represents a cutting-edge advancement in object detection, integrating attention mechanisms to enhance detection accuracy while maintaining the computational efficiency characteristic of previous YOLO iterations.

This guide delineates the procedural framework for configuring, training, and deploying YOLOv12 on a Windows-based system, encompassing installation protocols, dataset preparation, model training, and inference methodologies.

Step 1: Environment Configuration

Prior to initiating the setup, ensure that the following requisite software components are installed:

Essential Software and Tools:

• Python: Download and install Python (version 3.8 or later) from the official website, ensuring compatibility with YOLOv12 dependencies.
• Git: Install Git to facilitate repository cloning and version control.
• Integrated Development Environment (IDE): Utilize Visual Studio Code (VS Code) or an alternative IDE for script editing and execution.
• NVIDIA GPU Drivers: If deploying GPU acceleration, install the latest NVIDIA drivers to ensure optimal performance.
• CUDA and cuDNN: Required for GPU-based model training and inference.

Dependency Installation

Additional Dependencies: If utilizing external dataset management tools:

pip install roboflow supervision

Installing Python Dependencies: Execute:

pip install -r requirements.txt
pip install -e .

If errors occur, ensure setuptools and wheel are updated:

pip install --upgrade setuptools wheel

Cloning the YOLOv12 Repository: Execute the following commands in the terminal or command prompt:

git clone https://github.com/sunsmarterjie/yolov12.git
cd yolov12

Step 2: Dataset Preparation

YOLOv12 employs a dataset format akin to YOLOv8. The following steps ensure proper data structuring:

Dataset Formatting:

• YOLOv8 PyTorch TXT Format: Each image must have a corresponding .txt file with bounding box annotations.

Utilizing Roboflow for Dataset Management:

1. Creating a Roboflow Account: Register for a Roboflow account to facilitate dataset organization and download.
2. Dataset Conversion (if required): Convert non-YOLO formats (e.g., COCO JSON) using Roboflow utilities.
3. Creating data.yaml Configuration:

train: path/to/train/images
val: path/to/validation/images
test: path/to/test/images
nc: number_of_classes
names: ['class1', 'class2', ...]

4. Dataset Retrieval via API:

from roboflow import Roboflow

ROBOFLOW_API_KEY = "YOUR_API_KEY_HERE"
rf = Roboflow(api_key=ROBOFLOW_API_KEY)
project = rf.workspace("your-workspace").project("your-project")
version = project.version("your-version")
dataset = version.download("yolov8")

Step 3: Model Training

Executing Training Commands:

Utilize the following script, specifying the model configuration and dataset path:

from ultralytics import YOLO

model = YOLO('yolov12s.yaml')
results = model.train(
    data='path/to/your/data.yaml',
    epochs=250,
)

Real-World Implementation: Traffic Object Detection

model = YOLO('yolov12m.yaml')
results = model.train(
    data='traffic_data.yaml',
    epochs=300,
    batch=16,
    imgsz=640
)

This example demonstrates YOLOv12’s capability in detecting traffic-related entities such as vehicles, pedestrians, and signals.

Step 4: Inference Execution

Single-Image Inference:

from ultralytics import YOLO

model = YOLO('path/to/trained/model.pt')
results = model('path/to/image.jpg')
results.show()

Real-Time Video Inference:

import cv2
from ultralytics import YOLO

model = YOLO('yolov12s.pt')
cap = cv2.VideoCapture(0)

while cap.isOpened():
    ret, frame = cap.read()
    if not ret:
        break
    
    results = model(frame)
    cv2.imshow('YOLOv12 Detection', results.render()[0])
    if cv2.waitKey(1) & 0xFF == ord('q'):
        break

cap.release()
cv2.destroyAllWindows()

Step 5: Model Deployment

Trained YOLOv12 models can be integrated into various deployment frameworks. Below is an API deployment example using Flask:

Deploying YOLOv12 via Flask API:

from flask import Flask, request, jsonify
from ultralytics import YOLO
from PIL import Image
import io

app = Flask(__name__)
model = YOLO('yolov12s.pt')

@app.route('/detect', methods=['POST'])
def detect():
    image = Image.open(io.BytesIO(request.files['image'].read()))
    results = model(image)
    return jsonify(results.pandas().xyxy[0].to_dict())

if __name__ == '__main__':
    app.run(debug=True)

This API facilitates object detection by accepting image uploads and returning results in JSON format.

Conclusion

The implementation of YOLOv12 on Windows necessitates meticulous configuration, ranging from software installation to real-time inference and deployment.

By adhering to the outlined procedures, practitioners can effectively leverage YOLOv12’s capabilities for advanced object detection tasks, ensuring both accuracy and efficiency in real-world applications.

References

1. Run DeepSeek Janus-Pro 7B on Mac: A Comprehensive Guide Using ComfyUI
2. Run DeepSeek Janus-Pro 7B on Mac: Step-by-Step Guide
3. Run Microsoft OmniParser V2 on Ubuntu : Step by Step Installation Guide
4. Run YOLOv12 on Linux / Ubuntu: Step-by-Step Installation Guide